Integrating Multidimensional Scaling with Data Provenance Techniques
An interdisciplinary approach for assessing treatment motivation among patients undergoing antiretroviral therapy
This project is an interdisciplinary collaboration between the IIT DBGroup and Dr. Eric Houston from the IIT Psychology department. The overall objective of the study is to develop a novel, web-based application that will assess treatment motivation among individual patients living with HIV, and to explore the viability of its use in a clinical setting to promote medication adherence. The web-based application will collect data from patients to generate visual mappings of underlying conceptualizations of treatment that may interfere with their adherence and indicate poor treatment motivation. Current assessment approaches are unable to fully and accurately identify these conceptualizations. To aid healthcare providers in interpreting the visual mappings, models that automatically track the origins and processing of data (i.e., data provenance) will be created and embedded in the application used to generate these mappings.
This study has the following specific aims:
- To develop a technological platform that can be employed in clinical settings to obtain information on how individual patients conceptualize their treatment regimens and then use this information to generate multidimensional scaling (MDS) derived assessments of treatment motivation.
- To explore the feasibility and acceptability of utilizing this web-based assessment tool to improve treatment motivation and adherence among patients undergoing antiretroviral therapy (ART).
- To improve our understanding of data provenance for data mining algorithms such as MDS. As part of this aim, we will: a) address challenges such as finding the right model to express MDS provenance and determining how to automatically generate provenance information for MDS; and b) investigate whether the concepts developed for MDS translate to other data mining techniques.
Psychological Background and Multidimensional Scaling
Antiretroviral therapy (ART) for HIV/AIDS requires adherence rates of 95% or higher for a patient to achieve immunological benefits and virologic suppression. Research indicates, however, that 50% of patients are unable to meet these stringent standards.1 Patients who are not adherent at optimal levels are less likely to derive the immunological benefits of their regimens, and they risk developing treatment resistant strains of the virus which could be transmitted to others during sexual contact and drug use. The demands of ART regimens, the stigma associated with being HIV seropositive, and the effect of competing goals may strengthen treatment-related conceptualizations that lead to motivation for periodic lapses or complete avoidance of these regimens. Some conceptualizations may remain hidden to clinicians and healthcare providers until poor adherence behaviors have taken a toll on the patient’s health.
Multidimensional scaling (MDS) is an exploratory data analysis technique that can be employed to uncover latent conceptualizations prompting motivation to either engage or disengage from treatment for reasons that may be specific to the individual patient and not readily identifiable by clinicians.3,4 MDS provides information that current treatment motivation assessment approaches (i.e., self-report questionnaires) may be unable to elicit due to social desirability response bias and a tendency of these measures to tap only motivation to approach a health goal while largely overlooking the simultaneous effects of other competing forms of motivation. In addition, MDS provides assessment information that otherwise generally requires clinicians with training in specialized interviewing techniques.
Using a web-based application with provenance support, an MDS assessment could provide a thorough, timely and visually-enhanced assessment of patient treatment motivation. Such an assessment tool could be used to design patient-centered interventions aimed at promoting adherence. Thus, MDS has much potential to address an important public health concern. We propose an interdisciplinary study based on expertise from computer science and psychology to develop the application that patients can use in clinical settings via tablet devices or laptops. The proposed mixed-method study would strengthen our understanding and assessment of motivation, an important construct in psychology, and serve as the first step toward application for a R34 developmental grant aimed at addressing an important public health challenge posed by poor or suboptimal HIV medication adherence. In addition, the study would lead to the application and development of database provenance techniques from computer science for MDS that would aid in the generation and interpretation of assessments. There is a scarcity of research focusing on how database provenance techniques could be applied to data mining algorithms such as MDS.
Using Provenance to Interpret Data Mining Results - Application to MDS
For the user, it is critical to understand whether a phenomenon observed in the output of a data mining algorithm is based on the actual data or merely depends on the parameter choice and assumptions made by the mining algorithm. Furthermore, even though summarization is an important goal in data mining, it is important to be able to trace back a summarized result to the base data from which it is derived. In databases, such information about the creation process and origin of data is called data provenance. Traditional database provenance tracks on which input data a specific output of a query depends on, and how to automatically compute this information. In this study, we aim to develop provenance techniques for data mining algorithms. This is a challenging problem because the data processing of data mining algorithms is quite different from the database query processing, and thus requires novel provenance models and tracking methods. We will develop a model for MDS provenance and study how to generate provenance automatically for a MDS result. In addition, we will investigate how to represent this provenance information to the user. Finally, as part of the overall objectives of the study, we will explore whether we can automatically determine the extent to which results depends on the data versus the parameter settings.
- Eric Houston - Assistant Professor of Psychology at IIT, USA